Developers apply changes and updates to software systems to adapt to emerging environments and address new requirements. In turn, these changes introduce additional software defects, usually caused by our inability to comprehend the full scope of the modi ed code. As a result, software practitioners have developed tools to aid in the detection and prediction of imminent software defects, in addition to the eort required to correct them. Although software development eort prediction has been in use for many years, research into defect-correction eort prediction is relatively new. The increasing complexity, integration and ubiquitous nature of current software systems has sparked renewed interest in this eld. Eort prediction now plays a critical role in the planning activities of managers. Accurate predictions help corporations budget, plan and distribute available resources eectively and e ciently. In particular, early defect-correction eort predictions could be used by testers to set schedules, and by managers to plan costs and provide earlier feedback to customers about future releases. In this work, we address the problem of predicting the eort needed to resolve a software defect. More speci cally, our study is concerned with defects or issues that are reported on an Issue Tracking System or any other defect repository. Current approaches use one prediction method or technique to produce eort predictions. This approach usually suers from the weaknesses of the chosen prediction method, and consequently the accuracy of the predictions are aected. To address this problem, we present a composite prediction framework. Rather than using one prediction approach for all defects, we propose the use of multiple integrated methods which complement the weaknesses of one another. Our framework is divided into two sub-categories, Similarity-Score Dependent and Similarity-Score Independent. The Similarity-Score Dependent method utilizes the power of Case-Based Reasoning, also known as Instance-Based Reasoning, to compute predictions. It relies on matching target issues to similar historical cases, then combines their known eort for an informed estimate. On the other hand, the Similarity-Score Independent method makes use of other defect-related information with some statistical manipulation to produce the required estimate. To measure similarity between defects, some method of distance calculation must be used. In some cases, this method might produce misleading results due to observed inconsistencies in history, and the fact that current similarity-scoring techniques cannot account for all the variability in the data. In this case, the Similarity-Score Independent method can be used to estimate the eort, where the eect of such inconsistencies can be reduced. We have performed a number of experimental studies on the proposed framework to assess the eectiveness of the presented techniques. We extracted the data sets from an operational Issue Tracking System in order to test the validity of the model on real project data. These studies involved the development of multiple tools in both the Java programming language and PHP, each for a certain stage of data analysis and manipulation. The results show that our proposed approach produces signi cant improvements when compared to current methods.

Developers apply changes and updates to software systems to adapt to emerging environments and address new requirements. In turn, these changes introduce additional software defects, usually caused by our inability to comprehend the full scope of the modi ed code. As a result, software practitioners have developed tools to aid in the detection and prediction of imminent software defects, in addition to the eort required to correct them. Although software development eort prediction has been in use for many years, research into defect-correction eort prediction is relatively new. The increasing complexity, integration and ubiquitous nature of current software systems has sparked renewed interest in this eld. Eort prediction now plays a critical role in the planning activities of managers. Accurate predictions help corporations budget, plan and distribute available resources eectively and e ciently. In particular, early defect-correction eort predictions could be used by testers to set schedules, and by managers to plan costs and provide earlier feedback to customers about future releases. In this work, we address the problem of predicting the eort needed to resolve a software defect. More speci cally, our study is concerned with defects or issues that are reported on an Issue Tracking System or any other defect repository. Current approaches use one prediction method or technique to produce eort predictions. This approach usually suers from the weaknesses of the chosen prediction method, and consequently the accuracy of the predictions are aected. To address this problem, we present a composite prediction framework. Rather than using one prediction approach for all defects, we propose the use of multiple integrated methods which complement the weaknesses of one another. Our framework is divided into two sub-categories, Similarity-Score Dependent and Similarity-Score Independent. The Similarity-Score Dependent method utilizes the power of Case-Based Reasoning, also known as Instance-Based Reasoning, to compute predictions. It relies on matching target issues to similar historical cases, then combines their known eort for an informed estimate. On the other hand, the Similarity-Score Independent method makes use of other defect-related information with some statistical manipulation to produce the required estimate. To measure similarity between defects, some method of distance calculation must be used. In some cases, this method might produce misleading results due to observed inconsistencies in history, and the fact that current similarity-scoring techniques cannot account for all the variability in the data. In this case, the Similarity-Score Independent method can be used to estimate the eort, where the eect of such inconsistencies can be reduced. We have performed a number of experimental studies on the proposed framework to assess the eectiveness of the presented techniques. We extracted the data sets from an operational Issue Tracking System in order to test the validity of the model on real project data. These studies involved the development of multiple tools in both the Java programming language and PHP, each for a certain stage of data analysis and manipulation. The results show that our proposed approach produces signi cant improvements when compared to current methods.

Software project management is arguably the most important activity in modern software development projects. In the absence of realistic and objective management, the software development process cannot be managed in an effective way. Software development effort estimation is one of the most challenging and researched problems in project management. With the advent of object-oriented development, there have been studies to transpose some of the existing effort estimation methodologies to the new development paradigm. However, there is not in existence a holistic approach to estimation that allows for the refinement of an initial estimate produced in the requirements gathering phase through to the design phase. A SysML point methodology is proposed that is based on a common, structured and comprehensive modeling language (OMG SysML) that factors in the models that correspond to the primary phases of object-oriented development into producing an effort estimate. This dissertation presents a Function Point-like approach, named Pattern Point, which was conceived to estimate the size of object-oriented products using the design patterns found in object interaction modeling from the late OO analysis phase. In particular, two measures are proposed (PP1 and PP2) that are theoretically validated showing that they satisfy wellknown properties necessary for size measures. An initial empirical validation is performed that is meant to assess the usefulness and effectiveness of the proposed measures in predicting the development effort of object-oriented systems. Moreover, a comparative analysis is carried out; taking into account several other size measures. The experimental results show that the Pattern Point measure can be effectively used during the OOA phase to predict the effort values with a high degree of confidence. The PP2 metric yielded the best results with an aggregate PRED (0.25) = 0.874.

Software prediction unveils itself as a difficult but important task which can aid the manager on decision making, possibly allowing for time and resources sparing, achieving higher software quality among other benefits. One of the approaches set forth to perform this task has been the application of machine learning techniques. One of these techniques are Bayesian Networks, which have been promoted for software projects management due to their special features. However, the pre-processing procedures related to their application remain mostly neglected in this field. In this context, this study presents an assessment of automatic Bayesian Networks (i.e., Bayesian Networks solely based on data) on three public data sets and brings forward a discussion on data pre-processing procedures and the validation approach. We carried out a comparison of automatic Bayesian Networks against mean and median baseline models and also against ordinary least squares regression with a logarithmic transformation, which has been recently deemed in a comprehensive study as a top performer with regard to accuracy. The results obtained through careful validation procedures support that automatic Bayesian Networks can be competitive against other techniques, but still need improvements in order to catch up with linear regression models accuracy-wise. Some current limitations of Bayesian Networks are highlighted and possible improvements are discussed. Furthermore, this study provides some guidelines on the exploration of data. These guidelines can be useful to any Bayesian Networks that use data for model learning. Finally, this study also confirms the potential benefits of feature selection in software effort prediction.

Software prediction unveils itself as a difficult but important task which can aid the manager on decision making, possibly allowing for time and resources sparing, achieving higher software quality among other benefits. One of the approaches set forth to perform this task has been the application of machine learning techniques. One of these techniques are Bayesian Networks, which have been promoted for software projects management due to their special features. However, the pre-processing procedures related to their application remain mostly neglected in this field. In this context, this study presents an assessment of automatic Bayesian Networks (i.e., Bayesian Networks solely based on data) on three public data sets and brings forward a discussion on data pre-processing procedures and the validation approach. We carried out a comparison of automatic Bayesian Networks against mean and median baseline models and also against ordinary least squares regression with a logarithmic transformation, which has been recently deemed in a comprehensive study as a top performer with regard to accuracy. The results obtained through careful validation procedures support that automatic Bayesian Networks can be competitive against other techniques, but still need improvements in order to catch up with linear regression models accuracy-wise. Some current limitations of Bayesian Networks are highlighted and possible improvements are discussed. Furthermore, this study provides some guidelines on the exploration of data. These guidelines can be useful to any Bayesian Networks that use data for model learning. Finally, this study also confirms the potential benefits of feature selection in software effort prediction.

Software prediction unveils itself as a difficult but important task which can aid the manager on decision making, possibly allowing for time and resources sparing, achieving higher software quality among other benefits. One of the approaches set forth to perform this task has been the application of machine learning techniques. One of these techniques are Bayesian Networks, which have been promoted for software projects management due to their special features. However, the pre-processing procedures related to their application remain mostly neglected in this field. In this context, this study presents an assessment of automatic Bayesian Networks (i.e., Bayesian Networks solely based on data) on three public data sets and brings forward a discussion on data pre-processing procedures and the validation approach. We carried out a comparison of automatic Bayesian Networks against mean and median baseline models and also against ordinary least squares regression with a logarithmic transformation, which has been recently deemed in a comprehensive study as a top performer with regard to accuracy. The results obtained through careful validation procedures support that automatic Bayesian Networks can be competitive against other techniques, but still need improvements in order to catch up with linear regression models accuracy-wise. Some current limitations of Bayesian Networks are highlighted and possible improvements are discussed. Furthermore, this study provides some guidelines on the exploration of data. These guidelines can be useful to any Bayesian Networks that use data for model learning. Finally, this study also confirms the potential benefits of feature selection in software effort prediction.